Research On User Recruitment And Resource Scheduling In Edge-Cloud-Based Mobile Crowdsensing Systems

As the computation,storage,and communication capabilities of smart terminal devices have increased significantly,a new computing paradigm called mobile crowdsensing has attracted great attention from academia and industry.Mobile crowdsensing is to combine multiple mobile users carrying smart terminal devices through wireless networks to complete a large-scale sensing mission that is very difficult for a single user to cope with.Due to the advantages of low cost,universality,and high flexibility,mobile crowdsensing has a wide range of application prospects.Since the mobile crowdsensing system will involve large-scale data analysis when processing mobile users’ returned results,the platform will face mounting computation and storage pressure.Especially,when the number of task initiators increases,the platform’s situation will be more difficult.Thanks to the development of edge computing,we build an edge-cloud-based mobile crowdsensing system.The platform can process and transmit mobile users’ sensing data with the help of some servers at the edge of the network(called edge nodes).In the edge-cloud-based mobile crowdsensing system,the most important module is how to select some suitable mobile users from a large number of user candidates to complete the task(called user recruitment),and how to schedule the resources of the edge nodes to assist mobile users process and transmit their sensing data(called resource scheduling).This thesis will study the user recruitment and resource scheduling issues in the edge-cloud-based mobile crowdsensing system.First of all,we will focus on the user recruitment problem.Generally speaking,the platform always wants to select mobile users with strong sensing capabilities to complete tasks.However,in actual scenarios,the platform cannot know mobile users’sensing capabilities in advance,which is called "unknown user recruitment".We adopt the Combinatorial Multi-Armed Bandit(CMAB)model of the reinforcement learning field to formalize the learning process of users’ sensing capabilities,and then propose a CMAB-based unknown user recruitment mechanism.In this mechanism,in order to avoid that the platform might fall into a dilemma,that is,whether to always recruit users who had "good" performance before,or try some users who temporarily performed "bad" to explore the potentially best users,we first extend the basic UCB(Upper Confidence Bound)formula,so that the upper confidence bound of users’ sensing quality in the CMAB model can be expressed more accurately.Then,we adopt the max{} function to determine the sensing quality value for each task in each round.Based on this,we propose a sensing quality function for all tasks,and then design a greedy user recruitment strategy,that is,we always select the user who can increase the marginal value of the proposed sensing function the most.Extensive theoretical analysis and experimental simulations based on real-world datasets have proven that our proposed mechanism has good performance.On the other hand,in the process of recruiting users,mobile users’ sensing quality and recruitment cost values often involve their private information,so there always exists a dispute between the system and mobile users about whether the sensing quality/cost values should be public.In addition,the system must satisfy the specific requirements of the task initiator,that is,the final completion quality of each task is higher than a specific threshold.To solve this problem,we design a privacy-preserving user reruitment mechanism.In this mechanism,in order to recruit the suitable users with the minimum cost,while ensuring the initiator’s requirements about the completion quality of tasks,we first design a utility function that combines the task completion quality and the threshold constraint,and then propose a greedy user selection strategy,that is,we always select the user who can maximize the marginal value of the utility function per cost.At the same time,in order to protect mobile users’private information(sensing quality and cost)from leaking to others,we combine the secret sharing scheme into the proposed strategy to devise the secure user recruitment mechanism.We have proved the absolute security of the proposed mechanism under the semi-honest security model through rigorous theoretical analysis,and a large number of theoretical analysis and simulations have also verified the significant performance of this mechanism.In the edge-cloud-based mobile crowdsensing system,how to effectively schedule the limited resources in the edge nodes to assist in processing mobile user’s sensing data,so as to maximize the performance of the entire system,is challenging.For this problem,we propose an edge-cloud-oriented virtual machine resource scheduling mechanism.In this mechanism,in order to solve the(different)deadline constraints of mobile users,we first generate some virtual edge nodes,thereby transferring the three-layer system framework into a two-layer framework.Meanwhile,to handle the competitive relationship among mobile users and the strategic behaviors of users(that is,misrepresenting information to improve users’ own profits),we use an auction to allocate virtual machine resources in the edge nodes,which mainly includes the winning bid selection algorithm and the payment determination algorithm.Specifically,we model the winning bid selection problem as a many-to-one weighted bipartite graph matching problem with multiple 0-1 knapsack constraints,and then we always choose the edge with the largest weight to add to the allocation solution.For the payment determination algorithm,we design the calculation method of the critical payment value according to Myerson’s theorem,so as to ensure the truthfulness.We have verified the truthfulness,individual rationality,and computational efficiency of this mechanism through theoretical analysis and the experimental simulation based on real-world datasets.After mobile users complete the sensing tasks,they need to upload the final sensing results to the system platform.On the one hand,mobile users can transmit data through the cellular network,but the transmission cost is relatively high;on the other hand,users can use the communication capabilities of the edge nodes to complete the data transmission.The transmission cost of the edge nodes is relatively low,but the disadvantages are that the transmission capacity is limited and the data transmission service provided by edge nodes is a probabilistic event.For this problem,we design a data scheduling and transmission mechanism based on the resource-limited edge nodes,which can minimize the user’s transmission cost and meet the mobile user’s transmission deadline constraint.More precisely,we first model this problem as an optimization problem with multiple 0-1 knapsack constraints,where the knapsack is equivalent to an edge node with limited transmission capacity.Because the successful transmission of each data item through an edge node is a probabilistic event,we allow a data item to be scheduled to multiple edge nodes to increase the successful probability of transmission.In other words,all data will share a combinatorial probability optimization goal.Also,we have proved the efficiency of the proposed mechanism through a large number of theoretical analysis and experimental simulations based on real-world datasets.